ducanhdinh/Qwen1.5-MoE-A2.7B-MidAlign
Adapter vs Qwen/Qwen1.5-MoE-A2.7B. ducanhdinh/Qwen1.5-MoE-A2.7B-MidAlign is a LoRA adapter, not full weights: rank 8, on down_proj, gate, gate_proj, k_proj, o_proj, q_proj, up_proj, v_proj in 8 layers (1504 adapted matrices). It modifies Qwen/Qwen1.5-MoE-A2.7B at load time.